Khulna's Silent Ledger: The Signal Buried in the NCL's Unwritten Scorecards

Khulna's Silent Ledger: The Signal Buried in the NCL's Unwritten Scorecards

In November of last season, at the eighth over of the second session on the third day at the Sheikh Abu Naser Stadium in Khulna, I drew a small star in my notebook. I was sitting in the western corner of the gallery, the one vantage point from which the distance between a bowler's release point and the keeper's applause is roughly legible. Five deliveries in that over had gone outside square leg. Two were full-ish, short-of-a-length balls flat-batted past the bowler. One the keeper took one-handed, diving right. At stumps the scorecard recorded a single line: 8-1-34-1. Eight overs, one maiden, thirty-four runs, one wicket.

The lie lives inside that line. The bowler was a left-arm seamer under twenty-four, who had taken the new ball in the first three matches of the season and had never bowled more than thirteen overs in an innings. In Khulna he bowled twenty-one across two innings. In the first innings his economy was 2.8; in the second spell of the second innings it was 5.6. His bowling average for the match? 26.4 — top ten of the season.

The numbers were not lying; they were waiting for a better question.

Context: the cricket nobody records

The National Cricket League is the spine of Bangladesh's first-class cricket. Eight divisional sides — Khulna, Rajshahi, Dhaka, Dhaka Metro, Chattogram, Sylhet, Barishal, Rangpur — roll through it year after year, and almost the entirety of Bangladeshi first-class batting and bowling averages is manufactured there. Yet the ball-by-ball database for these matches amounts to a handful of hand-written spreadsheets, partial coverage, and the fixed scorecards of some matches that reach a website a day later, a fortnight later, or never.

I started coding by hand in the 2026-20 season. At first it was for myself, because the kind of analysis that gets done in the Mirpur press box never weighs an NCL spell the same way. Over six seasons I have typed more than 41,000 legal deliveries across 189 matches, from scorecards and, where they exist, ball-by-ball logs. There is no footage. There is no speed gun. There is no Hawk-Eye. Which means I could not measure the bowler's actual speed — only proxies: extra rate, share of short deliveries, session-level run rate, over-to-over strike rotation, timing of wicket falls.

The first sampling limit was written down before I ran the query: my dataset does not measure pace, it measures control. Three proxies for control — wides plus no-balls, an estimated beaten-or-left share, and the percentage of runs arriving in boundaries. When all three move together, the bowler's repeatability is breaking. When only one moves, the match context is changing.

In Khulna, I learned that silence is also a dataset.

Core: inside the spell, beyond the average

Look at NCL bowling averages and you are looking at a photograph of the sky. The clouds are beautiful. They do not tell you where the rain will fall. Across the 189 matches I coded, in 61 percent of fast bowlers the control proxies all moved together in the second spell of the second innings. By how much?

Season-wide, extras per over in a bowler's first spell: 0.19. Second spell: 0.41. Third spell: 0.46. Control error more than doubles from the first spell to the second, while the bowling average moves by only four to six percent. That gap is the story. The average shows the outcome; the spell shows the capacity.

The next question is whether that collapse is fatigue or batting depth. This is where I needed a control group. I split the sample: Group A, bowlers who sent down more than 25 overs in a match; Group B, fewer than 25. Both groups showed a rising extra rate in the second spell, but Group A rose by a factor of 1.9 and Group B by 1.4. Within Group A, those who bowled in both innings saw their third-spell run rate 1.7 runs higher than Group B in the same over band.

So there is a distinct fatigue signal. But the larger signal sits inside selection. The bowlers who bowl the most overs are usually the bowlers whose team has no alternative. In several of the eight NCL squads, the pace resources are thin enough that a number-eleven bowler must send down four or five spells across two innings. On the scorecard he is a workhorse. In the data he is a workhorse. But nobody records the control decay of his second spell, because the scorecard has no separate line for that spell.

Which produces my second, less comfortable conclusion: NCL pace workload management runs on triage — on who is least bad rather than who is most good. A bowler who is good in the first three matches gets more overs in the sixth; a bowler who is bad in the first three gets fewer, and the opportunity to learn is cut off. The fatigue signal and the selection signal obscure each other.

The archive bias: the matches that never become data

There is a settled narrative about first-class surfaces in Bangladesh — that they favour spin, so spin averages are not lying. I am not rejecting that. I want to know how evenly the archive recorded it.

Of the 189 matches, 54 finished in three days or fewer. Every one of those 54 has a complete ball-by-ball scorecard, because fewer days means less work and faster entry. Of the matches that ran five days, or that produced no result, 27 percent have either incomplete scorecards or a missing innings-by-innings over breakdown. The most complete part of the archive is the part made of the fastest-finished matches — which were almost universally spin-friendly.

This manufactures a statistical artifact I call Fast-Finish Bias. Five-day matches — where seamers bowl more, where the pitch holds, where innings run long — are underrepresented. The conclusion that spinners dominate the NCL is therefore partly cricket and partly a recording artifact.

The spike got spiked, but the pattern stayed in the data.

I want to be careful here. I am not claiming Khulna or Rajshahi pitches are not spin-friendly. Historically, through the mid-November to mid-January window, those venues take extra turn in the fourth innings, and two or three spinners can control the game's tempo together. I am saying that the evidence we cite for that claim is largely drawn from precisely the matches that finished early — which makes the reasoning circular.

The age curve we imported

I have an old objection to how Bangladesh uses its young seamers, and in these six seasons I found its quantitative footing.

International performance curves, especially for fast bowlers, are built mostly from SENA data — where childhood volume, pitch type, winter scheduling and workload management all differ. In that model a seamer's peak arrives at 27 to 30. In Bangladeshi domestic cricket I found something else: first-class seamers reach effective maturity roughly two to three years earlier, at 24 to 27, because their total ball volume is far lower, but so is their strength base and their recovery access. He bowls less, so he erodes less; but he bowls so little that his skill-acquisition cycle stays half-finished.

In my dataset, pace bowlers aged 23 to 26 who got more than 180 overs in a season improved their control proxies by an average of 11 percent the following season. Those who bowled fewer than 140 overs improved by 3 percent. The second group is the larger one, because teams do not give overs to a young bowler who has not proven himself, and the proven bowlers are often picked on batting.

A correction is required here. Although my sample is 189 matches, the count of unique pace bowlers is 74, and only 19 of them crossed 180 overs. Nineteen people can suggest a curve, but they cannot prove one; they can only bound it. I am writing that limit down, because the worst offence in domestic-cricket analysis is dressing a small sample in large words.

Whose ledger is it

Twenty years of covering domestic cricket keeps landing me on one technical problem: these matches have no permanence. An innings ends; the paper goes onto a clipboard; someone has to type it into a statistical database, if anyone does. In forty-degree heat, at venues without a press box, without internet, that entry frequently stops. A left-arm seamer's most informative spell — the one that might have changed his career — survives in one local reporter's notebook, if he does not throw it away.

I put it another way: the system needs its own ledger. An immutable, publicly verifiable, locally replicable record. The gap between what we want in cricket accounting and what we get is not technological, it is cultural. At Mirpur, a white-ball game yields every ball mapped, release positions logged, scientific metrics archived. In Khulna, at a first-class match in the same league, we sometimes do not record the minimum of second-innings field placements. What drains through the gap between those two realities is the actual baseline of Bangladeshi cricket, and it is unmeasurable by design.

In Khulna, I learned that silence is also a dataset — and we shrink that dataset every time we write a statistical note, because we assume silence is numerous when it merely has no footage.

Contrarian: correlation is not causation

This is my least comfortable thought. Everything above pushes toward one conclusion: heavy overs on young seamers correlates with their decline. But I understand the gap between correlation and cause.

If I had not written the hypothesis down first, retrospective storytelling would be easy now. Here is the accounting: the bowlers who bowled more overs were usually older, i.e. more experienced; the ones who bowled bad spells were usually loaded down because they were the squad's only alternative. And that lack of alternatives was a consequence of selection policy, not fatigue. The correlation is really a correlation between two separate practices — usage intensity and selection constraint.

There is another explanation I will not quietly set aside: sometimes a bad second spell is good batting strategy, not fatigue. When batsmen see a bowler's bounce drop, they wait, and waiting raises the run rate. There are traces of this in my data — if a spell's first two overs contain no beating ball, the run rate rises over the next three. The numbers are not the only explanation; the batsman's ability to read is maybe twenty percent of it.

I write these uncertainties down because of an old experience. In 2026 I joined a Dhaka startup as its first data hire and wrote a football analysis showing Abahani Limited Dhaka had scored 23 goals from 15.8 xG in their first 12 games. My editor spiked it: tactics talk is for the boys. Three weeks later the thesis proved true and it ran under a staff byline. That taught me that a correct number delivered at the wrong time is not a lie; it is simply unused.

It taught me something else I now write at the top: my analysis should first contain the numbers that argue against me. In this piece that is the nineteen-man sample — a curve leaning on a principle drawn from twenty-seven bowlers who have no fixed profile of their own.

I do not chase edges; I build a monastery around them.

Takeaway: the signal for the next round

So here is what I will watch, written down now so I do not have to explain my position later.

First: every seamer's extra rate over the first two overs of the second spell of the second innings. If it sits at one or two in two overs, I will not read it as bowler error. I will read it as session-level workload. Second: how much a home spinner's run rate rises between the seventh and twelfth overs, because that band is the least discussed and the most decisive character in domestic first-class cricket.

Third, a different texture: the duration, in minutes, of tail-end wickets. In the NCL there is a relationship between how long the tail bats and the team total, and that relationship is stronger in Khulna and Rajshahi than in Sylhet or Dhaka, because the pitch settles and the grass thins, and in that environment the quality of tail batting is often the truest card.

When people ask where data is destroying the beauty of the game, I write back: forty-three percent was not a gamble; it was a contract with variance. Before Russia 2026 I coded 1,240 goals across four years and published one checkable line — 43 percent of knockout goals would come from dead balls. The tournament delivered 73 from 169, or 43.2. But that result did not make me a prediction machine. The next year, running the same proxy on international football, I went toward a false verdict because I had changed the basis of the strategy while trying to make the number match.

The discipline is this: numbers must appear as upstream and downstream events, not merely as certificates for one another. Cricket is the same. This first-class league has been quietly teaching us that the glorious white-ball square at Mirpur and the pale scorecard of the NCL are one image, which we have split into two realities with a camera.

And on a morning when there is cricket in Khulna, I will take the seven o'clock bus, and the first thing I will look for is the notebook of the boy who wrote down the third spell behind the over-guard's back. If it turns up, we have a fragment of the ledger. If it does not, I will write the session myself — because every model is a prayer until the data says otherwise, and data does not arrive on its own. It has to be gathered by hand, in a silent stadium, in the afternoon heat, out of the middle of nothing happening.